Ophys data timestamps are not aligned with stimulus timestamps

[alexpiet]

@nicain Because the clocks of the stimulus presentation are different from the ophys sampling timestamps, I might sometimes want the "raw" data where I get ophys timestamps with the actual time. However, if I want to average across image presentations, I will want to align the ophys timestamps. I think there are two different ways to do this, and I think both are valid options!

1. I align the ophys timestamps by taking the closest timestamp relative to the stimulus onset.

2. I interpolate the ophys data to estimate the ophys signal on the stimulus onset.

[nicain]

@alexpiet Can you give a little more detail as to what the requested changes might be?

[alexpiet]

@nicain Just to make sure I was clear on the problem: the stimulus is presented are certain time points. The imaging takes place at a not-perfect multiple of the stimulus rate, meaning on some trials the first imaging data point will be exactly at time=0 when the stimulus is presented, but other trials the first imaging data point is some time after time=0.

If we want to look at neural responses aligned to stimulus presentation, the code in the visual behavior branch just grabs the nearest imaging point and calls it timepoint 1. But across trials timepoint 1 is slightly jittered. A better solution would be to always do linear interpolation to estimate the df/f signal at the same time point relative to each stimulus presentation.

This should be really easy to implement, but is probably very low priority.

[nicain]

@alexpiet OK, this makes sense, and we have worked on this issue before in the original passive-viewing experiment. Does this sound like a good summary:

"Feature request: Linear interpolation of dff timeseries onto the time grid defined by the timestamps from the stimulus_timestamps attribute."

@matchings @saskiad @mabuice @Gocker: Thoughts on this?

[alexpiet]

Your summary is perfect.

[matchings]

@nicain We may also want to have the ability to downsample the stimulus times onto the ophys time base, like what was done in Visual Coding. I can imagine users wanting any of these 3 options: use native timestamps and 'get nearest', use interpolated ophys timestamps and work in stimulus timebase, use downsampled stimulus timestamps and work in ophys timebase.

[saskiad]

I might suggest not having three versions of the alignment. I'd suggest having the true timestamps, and downsampling the stimulus to the ophys ala VIsual Coding. First, having three solutions generates confusions. Second, I find "get nearest" and interpolating ophys to be potentially problematic. If people really want to align the data that way, they always can if they have the true timestamps, but I'm reluctant to provide that as a default.

…

On Mon, May 13, 2019 at 6:40 PM Marina ***@***.***> wrote: @nicain <https://github.com/nicain> We may also want to have the ability to downsample the stimulus times onto the ophys time base, like what was done in Visual Coding. I can imagine users wanting any of these 3 options: use native timestamps and 'get nearest', use interpolated ophys timestamps and work in stimulus timebase, use downsampled stimulus timestamps and work in ophys timebase. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#635?email_source=notifications&email_token=AA5N4JSS2QTIIBNYAGDYRLLPVIKAFA5CNFSM4HL54TV2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODVKAOGY#issuecomment-492046107>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA5N4JTSHEUOL2QLQSEGRBLPVIKAFANCNFSM4HL54TVQ> .

[DowntonCrabby]

@saskiad @matchings @alexpiet do you all know if this was resolved, and what the solution was if so?

[matchings]

@DowntonCrabby This was not resolved in the SDK, but we do have temporal alignment functions in VBA (primarily used to make the response dataframes). Any functions for interpolation of data for alignment purposes should also live in VBA for now.

I don't think we are at the point where adding temporal alignment to the SDK is a high priority feature request. Once we have thoroughly vetted functions for temporal alignment / interpolation in VBA, and more clearly defined use cases for those functions, we can talk about moving them into the SDK. But we can get by with having them in VBA for now.

I would consider this issue "on hold". It might be best addressed as part of the larger Multiscope SDK infrastructure effort.